ddr 433 vs rdram 1066 and which board supports them

[Crucio]

I was wondering, is there any motherboard that is better than the P4T533-C? Or is it still the best for p4 533fsb and rdram.

Thankyou for your help!

[alexnoe]

Forget about DDR400 or DDR433. It doesn't work as supposed.

[theReal]

Forget about DDR400 or DDR433. It doesn't work as supposedWhat exactly doesn't work? I have no idea, but it'd be interesting to know in what ways it doesn't work as supposed.

[alexnoe]

Slower than DDR333 due to crappy timimgs, lack of any standard etc

[colordog]

I've read that as well in a bunch of on-line reviews about the DDR400 memory - it actually performs slower than the DDR333 memory. In a nutshell, while the memory transfer speed is faster (obvisously) than the DDR333, the DDR400 has bad latency. That means that there's a long wait between the really fast memory writes. When you average the fast writes out with the long waits, it's actually a bit slower than DDR333 memory. Sucks, eh? (I had been a big VIA fan up to this point). To make things worse, DDR400 isn't offically supported, because memorymakers and motherboard manufacturors couldn't decide on a common design! You'll actually see stickers on some motherboard boxes that say "Unofficially supports DDR400 memory!!!"
See "Why KT400 Does Not Support DDR400" at http://www.xbitlabs.com/news/story.html?id=1029904628
See one of many reviews where DDR400 is the same or worse than DDR333: http://www.anandtech.com/chipsets/showdoc.html?i=1660

Also, if you're thinking of P4 (in my opinion) you're going to want to say away from DDR - they don't feed the CPU fast enough. Alright though for the FSB of Althons.

I also saw you asking about the P4T533-C -- I just got two weeks ago the P4T533 (non-"C"), and I'd have to say I'm really happy with it. Lemme know if you have specific questions about it.

[alexnoe]

Dual Channel DDR (E7200 i think) is coming soon and will solve the problem of not feeding P4's fast enough ;)

[colordog]

Maybe, who can say? It doesn't exist yet, DDR400 isn't officially supported because VIA won't agree with memory makers... If you want the P4 to not be waiting on feeds from memory, go with the 1,066GHz memory Rambus, as much as I hate to say it....

[alexnoe]

There is no 1,066 GHz memory. What you are referring to is 16 Bit 133 MHz dual channel quad pumped or 32 Bit 133 MHz single channel quad pumped.

[colordog]

Sure there is, and I have some.

I have two 256MB sticks of 32bit RIMM4200 (aka PC1066). The motherboard utilizes the RIMMs via dual-channel rambus architecture. The FSB is quad-pumped to 533MHz. Since it's dual channel, the data rate operates at 1,066MHz. This translates to a theoretical possible bus width of 4.2GB/sec.

If you'd like to learn the all the details of how direct RamBus DRAM works, read this review, which explains "PC-1066 runs at 1066 MHz, which is the first consumer level memory technology to reach over 1 GHz in clock speed." It's a really cool article:
http://www.gamepc.com/labs/view_content.asp?id=stateofpc1066&page=1

Here is one of a zillion places that sells 1.066GHz RIMMs:
http://www.memorysuppliers.com/oczperser25p.html Don't buy it from them; there are better places.

[alexnoe]

The FSB is not 533, it is 133.

They just call it 533 (or 266 for Athlons) because it sounds better.
The frequency is *always* 133 (or 166). RD1066-RAM does nothing else than transfering 4 packets on each edge, giving 8 packets per cycle.

This frequency of 1,066 GHz is achieve nowhere.

[colordog]

I'm sorry that you're confused on this point, but the front side bus actually does run at 533MHz - this is not a marketing gimick - data is shuffled at that data rate. Just like Athlons XP run currently at 266 and soon/now at 333MHz, the Intel 850E chipset can run at a 533 MHz FSB. If you don't believe this, check any review of an Intel 850E board where they show screen shots of SandraSoft or WCPUID.

Additionally, you really should read the review concerning RIMM memory, as it should clear up your confusion. The internal clock really does run at 1,066 MHz = 1.066 GHz. (Not 1,066 GHz, which is a factor of a thousand error).

You may not want to believe that FSBs can operate at 533MHz, but they do, and you can buy them. There are dozens of reviews of them from various companies. Ditto for the memory. You might want to check things like this out on Google before simply claiming they don't exist without any basis or reference.

Here is one of the first reviews I found on Google where they talk about the 533 MHz FSB capabilities of this board complexed with 32bit RIMMs running at 1,066 GHz. Not my favorite, but one of the first that came up on Google.

http://www.hothardware.com/hh_files/Motherboards/asus_p4t533.shtml

The frequency is not always 133 MHz - you're probably confusing rambus with normal DDR. This isn't normal DDR. Check it out. Good luck!

[alexnoe]

The clock speed is 133. Nothing else. I don't know what verious reviews tell, since most reviewers just quote this from others who haven't understood it either. The only topic which is written even more crap about is SafeDisc 2 copy protection. One thing you have to learn ist that 90% of people who write reviews haven't understood the device they review.

The clock is 133 MHz, and 4 packages of data are transfered at each edge.
Since one cycle has 2 edges, 8 packages per cycle can be transfered.

Athlons transfer 1 package at each edge, thus giving 2 packages per cycle. AMD calls this "266 MHz", not because of 266 MHz, but due to 266 millions of edges per second. You confuse cycles, edges and data packages.

I know very well that WCPUID shows disinformation as well and refers to transfered packages instead of real clock speed.

About 1.066 vs 1,066: Please not that I mean the correct thing, however, German and US way of putting commas and dots is different ;)

These memory and boards have a throughput as if they indeed ran at 1066 MHz or 533 MHz or 266 MHz, but these are not the clock speeds.

[colordog]

Again, I'm sorry but I must insist that it's you who are confused. You simply dismiss information of referenced information, while having essentially nothing to backup your statements. Saying the clock speed is "133MHz and nothing else" and similar to saying that all processors also run at 133MHz and nothing else - but that's not true either. It's not just _solely_ increasing the number of bits or "packets" per clock cycle - the clock cycle itself increases. Certainly from a design point, it's much easier to engineer current databus bit widths at fast clock speeds than to keep increasing bit width every few months. That is, to increase speed, it's easier/cheaper to learn to build memory chips that read/write faster than it is to leave the memory clock and increase the bit width to 32, then 64, then 128, and soon you have to produce 1,024 bit memory chips to gain speed, the motherboards would have to be upgraded as soon, etc.

It is just as wrong to say that 32bit 1.066GHz memory in reality gets eight "packets" per 133MHz cycle as it is to say that a 2.4GHz Pentium 4 gets eighteen "packets" per 133MHz. The data width isn't set to eighteen, or set to the multiplier used. The write states in a Pentium 2.4GHz actually function as 2.4GHz, that is, they can change states at a theoretical maximum of 0.4 nanoseconds.

Calling these RIMMs "1.066GHz" are also true. The clock chip, on the MM itself, operates up to frequencies of 1.066GHz. This is in the specs. You cannot debate this, unless you claim that all manufacturors of these RIMMs and all RamBus are lying about how often their clock cycle increments forward. If you perform the easy math, you see that the theoretical mamximum write state (1 / 1.066^9) comes out to a little over 1 nanosecond. Of course, it's never that good in the real world (especially with mass manufactored chips), and in fact, 32bit 1,066GHz DDRamBus is rated at 32ns latency. Some people have had success overclocking that down to ~28ns.

But faster memory or faster CPUs actually CLOCK FASTER - it's not that a Pentium 2.4 has more "packet channels" than a Pentium 2.0.

If you have some source that states that the clock speeds throughout all components on the motherboard is all set to 133MHz, I'd really be interested to see it. But if you read the white papers of the RIMMs, you'll see that the internal clock speed of these chips are at 1.066GHz. Memory can be set through EACH individual pin of the RIMM at 1.066GHz. Don't confuse these with PCI bus transfer or simple DDR memory - this is not the same. This memory is already dual-channeled.

(1) Here's the tech paper on this memory - only the first (or second) page is revelevant where they define the bus speed:
http://www.rambus.com/downloads/32BitRIMM_w256Mb_d_0114_V0_9.pdf

(2) This is technical but explains the separation of its clock from PCI: http://www.rambus.com/downloads/RIMM3264_06.12.pdf

(3) What is your definition of a "packet"? A 32-bit register?

(4) On what basis do you make the claim that WCPUID "shows disinformation"?

(5) On what basis do you make the claim that the FSB operates at a frequency other than 400MHz or 533MHz as stated by CPU, motherboard, and memory manufactors, as well as reveiwers, and programs such as WCPUID and SandraSoft?

(6) On what basis do you make the claim that the memory clock speed operates at a frequency other than 1.066 GHz, as stated by motherboard, and memory manufactors, as well as reveiwers, memory design consortiums, and programs such as WCPUID and SandraSoft?

Nutshell: the only way to get THINGS from PLACE A to PLACE B faster is either (1) send out more things at once (bit width) and/or (2) send out more things at a faster rate. That's it.

Do you realize what the bitwidth would have to be for a Pentium 4 clocked at 3.06 "relative" GHz if "truly" clocked at 133MHz?!? Or likewise, for 1.066GHz memory to 'appear' to run at 1.066GHz when it 'truly' runs at 133MHz? It would require thousands of pins!

Really, read this stuff.

[alexnoe]

Ah, you were refering to the *internals* of RD-RAM...then it's OK.

(3) Not necessarily 32 Bit. Whatever width the corresponding bus has. For RD-RAM, it should be 32 Bit (would be 8x133x4 = 32x133 = about 4,2 GB/s...fits perfectly ;) )

(4)+(5) Have you ever heard of "quadpumped" architecture?

http://dch-faq.de/kap02.html
This is not made by people who want to sell something or want to become as known as possible, but by people who want to give information. Use babelfish in case you need it.
In the middle of this you find
[quote] P4-Willa - Pentium 4 Willamette, Sockel 423 und 478, FSB 100MHz ("400MHz QDR")
P4-North - Pentium 4 Northwood, Sockel 478, FSB 100MHz ("400MHz QDR")
P4-Nw133 - Pentium 4 Northwood, Sockel 478, FSB 100MHz und FSB 133MHz ("533MHz QDR")These tools show the "QDR FSB", just because people want to see this one.

(6) I was refering to the external memory clock, the one which is used by the chipset to communicate with the memory.